For conditions that don't generalize. If it would be a "mathematical law", spaced repetition would not work, because the retention time would be that of the last presentation. So, we've got some exception. Interference, reinforcement and generalization are another known influences.
Psychology doesn't have a good model of how memory works, whereas there are fairly detailed approximations to model the behavior of e.g. gases. So it isn't mathematics, and it doesn't approach the standards of physics.
And that's also part of what's wrong with cogsci. It has held physics too long as the standard to emulate.